[U-Boot] Regression in bootcmd handling in v2015.04-rc3?

[Stephen Warren <swarren@wwwdotorg.org>]

On 03/11/2015 02:20 PM, Karsten Merker wrote:
> On Wed, Mar 04, 2015 at 08:53:37PM +0100, Karsten Merker wrote:
>> On Tue, Mar 03, 2015 at 06:31:37PM -0500, Tom Rini wrote:
>>
>>> I've pushed v2015.04-rc3 out to the repository and tarballs should exist
>>> soon.
>> [...]
>>> As always, if anything is broken please speak up.
>>
>> Hello,
>>
>> I have tested v2015.04-rc3 and have stumbled over a problem with the
>> distro_bootcmd framework when trying to boot from a specific device
>> at the u-boot command prompt by using "run <device>_boot".
>> Autobooting without user interruption of the boot process works fine:
...
>> Interrupting the boot sequence and then manually executing e.g. "run
>> usb_boot" results in a "** Bad device specification usb devplist **"
>> error message:
>>
>> sunxi# run usb_boot
>>
>> USB device 0: Vendor: Intenso  Rev:      Prod: Rainbow
>>              Type: Removable Hard Disk
>>              Capacity: 7620.0 MB = 7.4 GB (15605760 x 512)
>> ** Bad device specification usb devplist **
>> sunxi#
>>
>> The reason for this is that in this constellation ${devnum} is
>> undefined, which leads to a failure of the "part list ${devtype}
>> ${devnum} devplist" command in ${scan_dev_for_boot_part}.  Older
>> versions of u-boot did not use the part list command, so this problem
>> did not occur there.  Using "run bootcmd_usb0" instead of "run
>> usb_boot" works, but my understanding (which may of course be wrong)
>> was that the user is expected to use "run <device>_boot" for booting
>> from a specific device at the u-boot command prompt, and that has
>> worked without problems in previous u-boot versions.
>
> Hello,
>
> as my original email appears to have fallen through the cracks,
> I'd like to ask again whether using "run <device>_boot" is
> "officially" expected to work or not.  If it is not, what is the
> officially supported way for a user to boot from a certain device
> at the u-boot prompt?  Should people use "setenv boot_targets
> foo; run bootcmd" instead?

Sorry, I didn't see the original email.

For manual booting, I would expect to always run the device-specific 
command (e.g. bootcmd_usb0 or bootcmd_mmc1) rather than the 
device-type-specific command (e.g. usb_boot, mmc_boot).

I'm not sure how the type-generic commands ever worked without variables 
such as ${devnum} set, since they all pass ${devnum} to a variety of 
commands in addition to the new "part list". Perhaps those commands 
default to devnum==0 if the parameter is missing, and you never wanted 
to boot e.g. from mmc 1 instead of mmc 0?